How to Tune and Extract Higher Performance with MVAPICH2 Libraries

Presenter(s): Dhabaleswar K.(DK) Panda (Ohio State)

Presentation Slides

The Ohio State University MVAPICH2 libraries support the latest MPI 3.1 standard and deliver high performance, scalability and fault tolerance for high-end computing systems using InfiniBand, Omni-Path, 10-40 GigE/iWARP and RoCE (V1 and V2) networking technologies. MVAPICH2-GDR library uses novel designs to exploit the cutting-edge GPUDirect technology to provide high performance for MPI applications on systems with NVIDIA-GPUs. These libraries have multiple features, parameters and knobs to optimize the performance on modern systems. However, many users are not fully-aware of all these features, optimization and tuning techniques. This talk is aimed to address these concerns and provide a set of concrete guidelines to XSEDE users to boost performance of their applications. We will start with an overview of the MVAPICH2 libraries and their features and optimized designs. Next, we will provide an in-depth overview of the runtime optimizations and tuning flexibility. We will demonstrate how you can tune and optimize these libraries to fit the needs of your application on a given system. Using a set of `Best Practice' examples, we will highlight the impact of tuning and optimizations on a set of common XSEDE applications including Amber, Lulesh, Hoomdblue, and MILC.

Bio --------- DK Panda is a Professor and University Distinguished Scholar of Computer Science and Engineering at the Ohio State University. He has published over 350 papers in the area of high-end computing and networking. The MVAPICH2 (High Performance MPI and PGAS over InfiniBand, iWARP and RoCE) libraries, designed and developed by his research group (http://mvapich.cse.ohio-state.edu), are currently being used by more than 2,550 organizations worldwide (in 79 countries). More than 360,000 downloads of this software have taken place from the project's site. This software is empowering several InfiniBand clusters (including the 10th, 13th and 25th ranked ones) in the TOP500 list. The RDMA packages for Apache Spark, Apache Hadoop and Memcached together with OSU HiBD benchmarks from his group (http://hibd.cse.ohio-state.edu) are also publicly available. These libraries are currently being used by more than 160 organizations in 22 countries. More than 15,900 downloads of these libraries have taken place. He is an IEEE Fellow. More details about Prof. Panda are available at http://www.cse.ohio-state.edu/~panda